This paper describes our work in progress on FlexCache, a framework for flexible, compiler generated data caching. FlexCache substitutes the tag-memory and cache controller hardware with a compiler managed tag-like data structure, address translation, and tag-check. This allows the division of the data-array into separately controlled partitions, allows the selection of cache line sizes, the support of highly associative mappings, and the selection of various replacement policies on a per program basis. FlexCache leverages compile-time static information to selectively virtualize memory, to eliminate cache-tag accesses, and to guide the replacement of conflicting cache lines. For the applications studied the FlexCache compiler techniques eliminate in average more than 90% of the cache-tag lookups, enabling the support of highly associative caching schemes. Even without any hardware support, FlexCache can outperform fixed hardware caches by improving caching effectiveness, eliminating mapping conflicts, and eliminating the cache pollution caused by register spills. The beauty of FlexCache is that its core techniques could be augmented with additional software techniques and/or hardware support (i.e., special instructions) to look into optimizing data caching in areas such as low-power, real-time systems, and high-performance microprocessors.
CITATION STYLE
Moritz, C. A., Frank, M. I., & Amarasinghe, S. (2001). FlexCache: A framework for flexible compiler generated data caching. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 2107, pp. 135–146). Springer Verlag. https://doi.org/10.1007/3-540-44570-6_9
Mendeley helps you to discover research relevant for your work.